Fig. 8
From: Molecular complex detection in protein interaction networks through reinforcement learning
Evaluating the predictions of the RL algorithm on hu.MAP 1.0. A The optimal Qi threshold is 0.325. We tested various overlap thresholds, i.e., Qi values (Eq. 7) between 0.2 and 0.9 in 0.25 intervals. B Size distributions of known and predicted complexes. This graph shows that the distribution of the sizes (no. of proteins) of the predicted and known complexes is very similar. C F1 score distributions of the best-predicted match for known complexes and vice-versa. In both cases, higher F1 scores have higher counts indicating accurate predictions